Update README.md

README.md

@@ -73,11 +73,14 @@ between objects** in a scene, rather than only their semantic identity.
 
 - 📄 **Paper:** [arXiv:2601.11729](https://arxiv.org/abs/2601.11729)
 - 💻 **Code:** [github.com/gmum/SpaRRTa](https://github.com/gmum/SpaRRTa)
-- 🧱 **Real-world (lego) split:** hosted separately at `turhancan97/SpaRRTa-Lego`
-
-This repository hosts the **synthetic (Unreal Engine 5)** portion of the benchmark. The
-complementary real-world set — photographed with toy minifigures for sim-to-real evaluation —
-is released at [`turhancan97/SpaRRTa-Lego`](https://huggingface.co/datasets/turhancan97/SpaRRTa-Lego).
+- 🧱 **Real-world (lego) split:** [turhancan97/SpaRRTa-Lego](https://huggingface.co/datasets/turhancan97/SpaRRTa-Lego)
+- 🔬 **Attention-analysis split (images + masks):** [turhancan97/SpaRRTa-Attention](https://huggingface.co/datasets/turhancan97/SpaRRTa-Attention)
+
+This repository hosts the **synthetic (Unreal Engine 5)** portion of the benchmark. It is
+complemented by two companion splits: the real-world set photographed with toy minifigures for
+sim-to-real evaluation ([`turhancan97/SpaRRTa-Lego`](https://huggingface.co/datasets/turhancan97/SpaRRTa-Lego)),
+and an attention-analysis set with per-object segmentation masks
+([`turhancan97/SpaRRTa-Attention`](https://huggingface.co/datasets/turhancan97/SpaRRTa-Attention)).
 
 <p align="center">
   <img src="assets/teaser.png" alt="SpaRRTa teaser" width="100%">